Two Approaches to Design of Monitoring Networks

Two different approaches to the design of a ground-water-quality monitoring network may be appropriate, depending on the type of information desired. First, where the objective is to determine what ground-water-quality characteristics are like in an area (statistical quantification of typical concentrations, as given by the mean or median, or percentage of wells exceeding various use standards), networks can be designed to provide estimates of known reliability using standard parametric and nonparametric statistical techniques. This approach can provide information adequate to perform general water-quality assessments where the intention of the monitoring network is to provide data about general suitability of the water for various uses. Second, where the objective is to maximize areal ground-water-quality information, networks also can be designed using geostatistical techniques, such as kriging. This second approach would be appropriate when information is needed on where particular problem areas may exist. Both approaches were applied to 1965 chloride data from a deep confined aquifer in the Llobregat delta near Barcelona, Spain. Traditional statistical techniques are demonstrated to design a network that would provide an estimated median chloride concentration. A method is introduced that can be used to determine the sample size necessary to describe any selected quantile with known precision. On the basis of 120 observations in the 1965 data set, between 13 and 25 wells would be necessary to estimate the median chloride concentration within 40 percent of the true median with 95-percent confidence. Kriging, a geostatistical technique, was applied to the data set to determine the minimum number of wells necessary to include in the network to retain the essential spatial information of the original network. By use of this technique, the original network of 120 wells was reduced by 17.5 percent to 99 wells, while the standard error was increased by only 1 percent. A comparison of these two approaches indicates that a network designed by use of geostatistical techniques generally will require larger sample sizes than networks designed by use of traditional techniques, but the geostatistical techniques can provide data adequate to describe both stochastic and spatial features of water-quality variables. Detailed description of spatial variability requires many sample points for extremely variable data. On the basis of results presented in this paper, prediction errors for chloride concentrations in ground water at selected points in the Llobregat delta were as much as 300 percent. Nevertheless, even the modified network of 99 wells would produce statistical estimates adequate for most general water-quality assessments, in addition to retaining the spatial information contained in the original 1965 data set.